OpenWhispr/openwhispr

OpenWhispr is worth a look if you type into other apps all day and want a single hotkey that turns speech into pasted text, but you're not comfortable sending everything to a hosted transcription service. It bundles local whisper.cpp and sherpa-onnx models alongside cloud options, so the same app covers dictation, meeting capture and notes.

The problem

Dictating into Slack, Gmail, an IDE or a chat window usually means opening a separate transcription page, speaking, copying the result and pasting it where you were working. Meetings add another step: record, then hunt for a transcript that a vendor may keep on its servers. OpenWhispr changes the middle step. A global hotkey records speech and pastes the cleaned text directly into whatever field has focus, and meeting recording can start automatically when Zoom, Teams or FaceTime calls are detected, producing a transcript with speaker labels.

What's different

The interesting mechanism is that transcription runs through the same pipeline whether the model is local or remote. Local options include Nvidia Parakeet and Whisper via whisper.cpp and sherpa-onnx; cloud providers are bring-your-own-key through OpenAI, Anthropic, Gemini, Groq, Tinfoil and OpenRouter, plus self-hosted transcription against custom ASR endpoints. That matters because it lets a privacy-conscious user keep audio on the machine while a user with a weak laptop can borrow a cloud model's accuracy.

The trade-off is scope. This is not just a dictation utility: it also ships an AI agent hotkey, notes with semantic search and folders, team spaces with sharing and roles, calendar integration, mobile app and an MCP server. Each adds integrations to configure and failure points to diagnose. Local speaker diarization is advertised as fully on-device, but accuracy claims for any model are the project's own and not independently verified here.

Who should try it

A good fit is someone who dictates across several apps on macOS, Windows or Linux and wants local models for sensitive content, with the option to switch to a cloud key when speed or accuracy matters. It also suits people who want meeting transcripts and notes in one place rather than juggling a recorder and a separate notes app.

A poor fit is someone who mainly needs a minimal, single-purpose dictation tool: the notes, team spaces, calendar sync and agent features are extra surface area you'd still carry. Anyone who only works on mobile should check the separate mobile README first, because the desktop root project and the Expo app have separate dependencies and release processes.

Requirements and limitations

From source, the desktop app needs Node.js 24+ and the usual npm install/npm run dev cycle; platform-specific setup is in the platform guides. Expect the first-run experience to involve picking a model path, choosing a provider and granting accessibility or input permissions so the global hotkey can paste into other apps — details vary by OS.

Ongoing upkeep is the bigger constraint. Local models must be downloaded and kept compatible with the bundled whisper.cpp/sherpa-onnx stack; cloud models need API keys and billing that you manage. The meeting detector, calendar integration and team sync each bring their own service configuration. If something misbehaves, the troubleshooting docs are the first place to look rather than the repository alone.

HotRepo's take

HotRepo only observed this repository on two days: 9,894 stars on 2026-10-24 (+57) and 9,907 on 2026-10-25 (+62). There is not enough comparable daily-board history to judge whether that interest persists, so treat the star count as a snapshot rather than a trend. Before installing, check the quickstart model guidance against your hardware — the local path is the part most likely to change your experience.